ROC Curve Calculator for Machine Learning

Calculator inputs

Calculation mode

Decimal precision

Metric display

Actual class labels

1,1,0,1,0,0,1,0,1,0,1,0

Use commas, spaces, tabs, or new lines.

Prediction probabilities or decision scores

0.95,0.88,0.81,0.76,0.70,0.62,0.59,0.44,0.41,0.35,0.28,0.12

Values outside 0–1 are accepted as decision scores.

Positive-class label

Positive score direction

Threshold generation

Threshold step size

Bootstrap repetitions

Random seed

Calculate percentile bootstrap AUC confidence interval

ROC coordinates

0,0,1.00
0.10,0.40,0.80
0.20,0.70,0.60
0.35,0.85,0.40
0.55,0.95,0.20
1,1,0.00

Enter one row per point. Format: FPR, TPR, optional threshold.

Threshold confusion matrices

0.90,2,0,6,4
0.70,4,1,5,2
0.50,5,2,4,1
0.30,6,4,2,0
0.10,6,6,0,0

Enter one row per threshold. Format: threshold, TP, FP, TN, FN.

Actual class labels

1,1,0,1,0,0,1,0,1,0,1,0

Model score rows

Model A|0.95,0.88,0.81,0.76,0.70,0.62,0.59,0.44,0.41,0.35,0.28,0.12
Model B|0.89,0.78,0.75,0.73,0.65,0.60,0.55,0.48,0.42,0.39,0.33,0.20

Format each row as Model name|comma-separated scores.

Positive-class label

Actual class labels

cat,dog,bird,cat,dog,bird,cat,dog,bird

Class names in probability-column order

cat,dog,bird

Probability matrix

0.85,0.10,0.05
0.10,0.80,0.10
0.10,0.15,0.75
0.70,0.20,0.10
0.20,0.65,0.15
0.10,0.20,0.70
0.75,0.15,0.10
0.15,0.70,0.15
0.15,0.10,0.75

One observation per row. Columns must follow the class-name order.

Threshold optimisation and operating costs

Optimisation method

Minimum sensitivity

Minimum specificity

Custom threshold

Deployment prevalence

False-positive cost

False-negative cost

True-positive benefit

True-negative benefit

Confidence level

Import CSV

Formula used

True Positive Rate (Sensitivity) = TP ÷ (TP + FN)
False Positive Rate = FP ÷ (FP + TN)
Specificity = TN ÷ (TN + FP) = 1 − FPR
Precision = TP ÷ (TP + FP)
Balanced Accuracy = (Sensitivity + Specificity) ÷ 2
Youden’s J = TPR − FPR
Top-left Distance = √[FPR² + (1 − TPR)²]
AUC = trapezoidal area beneath the ROC curve
Gini Coefficient = 2 × AUC − 1

A ROC curve plots sensitivity against false positive rate. Each point represents one decision threshold. AUC measures how well scores rank positive cases above negative cases.

How to use

Select the calculation mode matching your available data.
Paste labels, scores, coordinates, matrices, or model rows.
Select the positive class and score direction.
Choose a threshold objective and optional operating constraints.
Enter misclassification costs when errors have different consequences.
Enable bootstrap analysis when an empirical AUC interval is needed.
Submit the form, inspect the ROC graph, and review warnings.
Export the summary, table, graph, JSON, CSV, or PDF report.

Example data

Observation	Actual label	Prediction score	Meaning
1	1	0.95	Likely positive case
2	1	0.88	Likely positive case
3	0	0.81	Possible false positive
4	1	0.76	Moderate positive score
5	0	0.35	Likely negative case

Frequently asked questions

What does a ROC curve show?

It shows sensitivity and false positive rate across decision thresholds.

What does AUC mean?

AUC measures ranking discrimination across all available thresholds.

Is a higher AUC always better?

Usually, but deployment costs and threshold performance still matter.

Why can AUC fall below 0.50?

The positive label or score direction may be reversed.

Which threshold should I choose?

Choose one matching sensitivity, specificity, costs, and operational limits.

What is Youden’s J?

It equals sensitivity plus specificity minus one.

When is a precision-recall curve preferable?

It is often more informative for strongly imbalanced positive classes.

Does ROC measure probability calibration?

No. ROC measures discrimination, not probability calibration quality.

Can multiclass predictions use ROC curves?

Yes. One-vs-rest curves produce class and average AUC values.

Calculator inputs

Threshold optimisation and operating costs

Formula used

How to use

Example data

Frequently asked questions

What does a ROC curve show?

What does AUC mean?

Is a higher AUC always better?

Why can AUC fall below 0.50?

Which threshold should I choose?

What is Youden’s J?

When is a precision-recall curve preferable?

Does ROC measure probability calibration?

Can multiclass predictions use ROC curves?

Related Calculators